Fine-tuning Mistral to clone your Telegram persona
A four-stage pipeline to extract Telegram chat history, tokenize it, fine-tune Mistral 7B with LoRA, and deploy an interactive agent that talks like you.
AI & Systems Engineer
I build AI systems and write about the craft. From LLM fine-tuning and data pipelines to production infrastructure — I care about depth, clarity, and systems that actually ship.
Skills & Stack
Designing, training, and serving large language models. End-to-end ML pipelines: data extraction, tokenization, LoRA fine-tuning, and GPU inference — with a focus on reproducibility and production-readiness.
Primary language for ML systems, data processing, pipeline orchestration, and scripting.
Building robust ETL flows for ML datasets — from raw exports to clean, tokenized training data at scale.
Public tooling and experiments — code that ships, not prototypes collecting dust.
You2AgentAI ↗Static sites and tooling with Astro + Tailwind on Cloudflare Pages. Type-safe, zero-runtime where possible.
Recent writing